ceph: do not confuse stale and dead (unreconnected) caps
We were using the cap_gen to track both stale caps (caps that timed out
due to temporarily losing touch with the mds) and dead caps that did not
reconnect after an MDS failure. Introduce a recon_gen counter to track
reconnections to restarted MDSs and kill dead caps based on that instead.
Rename gen to cap_gen while we're at it to make it more clear which is
which.
Signed-off-by: Sage Weil <sage@newdream.net>
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index 8b863db..775e6f6 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -609,7 +609,8 @@
cap->seq = seq;
cap->issue_seq = seq;
cap->mseq = mseq;
- cap->gen = session->s_cap_gen;
+ cap->cap_gen = session->s_cap_gen;
+ cap->recon_gen = session->s_recon_gen;
if (fmode >= 0)
__ceph_get_fmode(ci, fmode);
@@ -626,17 +627,25 @@
static int __cap_is_valid(struct ceph_cap *cap)
{
unsigned long ttl;
- u32 gen;
+ u32 gen, recon_gen;
spin_lock(&cap->session->s_cap_lock);
gen = cap->session->s_cap_gen;
+ recon_gen = cap->session->s_recon_gen;
ttl = cap->session->s_cap_ttl;
spin_unlock(&cap->session->s_cap_lock);
- if (cap->gen < gen || time_after_eq(jiffies, ttl)) {
+ if (cap->recon_gen != recon_gen) {
+ dout("__cap_is_valid %p cap %p issued %s "
+ "but DEAD (recon_gen %u vs %u)\n", &cap->ci->vfs_inode,
+ cap, ceph_cap_string(cap->issued), cap->recon_gen,
+ recon_gen);
+ return 0;
+ }
+ if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
dout("__cap_is_valid %p cap %p issued %s "
"but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
- cap, ceph_cap_string(cap->issued), cap->gen, gen);
+ cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
return 0;
}
@@ -2203,7 +2212,8 @@
issued = __ceph_caps_issued(ci, &implemented);
issued |= implemented | __ceph_caps_dirty(ci);
- cap->gen = session->s_cap_gen;
+ cap->cap_gen = session->s_cap_gen;
+ cap->recon_gen = session->s_recon_gen;
__check_cap_issue(ci, cap, newcaps);